⚡ Quick Answer
A LoRA (Low-Rank Adaptation) is a small file that adjusts an existing checkpoint toward a specific style, character, or subject — without replacing the checkpoint itself. Think of it as a style filter you stack on top of your base model.
Training a full checkpoint from scratch takes huge amounts of data and GPU time. A LoRA does something smaller and cheaper: it learns one specific concept — a face, an art style, a clothing type — and applies it as a lightweight adjustment on top of a checkpoint you already have.
Why Use a LoRA
- Small file size — usually 20MB to 200MB, versus gigabytes for a checkpoint
- Fast to train and fast to load
- Stackable — you can load multiple LoRAs in one workflow to combine styles
- Doesn't touch your base checkpoint — remove it and the checkpoint behaves exactly as before
Where You'll See It
The LoraLoadernode sits between your Load Checkpoint node and the rest of your workflow. It takes the MODEL and CLIP outputs from your checkpoint, applies the LoRA's adjustments, and passes modified versions downstream.
Quick Example
If you load flux1-dev.safetensors as your checkpoint and add a LoRA trained on watercolor paintings, your outputs shift toward a watercolor look — while the underlying FLUX model still handles prompt understanding and image quality the same way it always did.
Stacking Multiple LoRAs
You can chain several LoraLoader nodes to combine effects — one for style, one for a specific character. Each LoRA has a strength setting (usually 0–1) that controls how strongly it affects the output.
Frequently Asked Questions
See It In Action
Ready to stack your first LoRA?
Our full guide walks through loading, stacking, and tuning LoRA strength step by step.
Published: 2026-08-25 · Last updated: 2026-08-25
Join the discussion
Sign in to leave a comment or reply
No comments yet
Be the first to share your thoughts!